Gender Bias in Language

Artificial Intelligence (AI) has become ubiquitous in human society, and yet vast segments of the global population have no, little, or counterproductive information about AI. It is necessary to teach AI topics on a mass scale. While there is a rush to implement academic initiatives, scant attention has been paid to the unique challenges of teaching AI curricula to a global and culturally diverse audience with varying expectations of privacy, technological autonomy, risk preference, and knowledge sharing. Our study fills this void by focusing on AI elements in a new framework titled Culturally Adaptive Thinking in Education for AI (CATE-AI) to enable teaching AI concepts to culturally diverse learners. Failure to contextualize and sensitize AI education to culture and other categorical human-thought clusters, can lead to several undesirable e ects including confusion, AI-phobia, cultural biases to AI, increased resistance toward AI technologies and AI education. We discuss and integrate human behavior theories, AI applications research, educational frameworks, and human centered AI principles to articulate CATE-AI. In the first part of this paper, we present the development a significantly enhanced version of CATE. In the second part, we explore textual data from AI related news articles to generate insights that lay the foundation for CATE-AI, and support our findings. The CATE-AI framework can help learners study artificial intelligence topics more e ectively by serving as a basis for adapting and contextualizing AI to their sociocultural needs.

Introduction: Post-editing plays a crucial role in enhancing the quality of machine-generated translation (MGT) by correcting errors and ensuring cohesion and coherence. With advancements in artificial intelligence, Large Language Models (LLMs) like ChatGPT-4o offer promising capabilities for postediting tasks. This study investigates the effectiveness of ChatGPT-4o as a natural language processing tool in post-editing Arabic translations across various domains, aiming to evaluate its performance in improving productivity, accuracy, consistency, and overall translation quality. Methods: The study involved a comparative analysis of Arabic translations generated by Google Translate. These texts, drawn from multiple domains, were post-edited by two professional human translators and ChatGPT-4o. Subsequently, three additional professional human post-editors evaluated both sets of post-edited outputs. To statistically assess the differences in quality between humans and ChatGPT-4o post-edits, a paired t-test was employed, focusing on metrics such as fluency, accuracy, coherence, and efficiency. Results: The findings indicated that human post-editors outperformed ChatGPT-4o in most quality metrics. However, ChatGPT-4o demonstrated superior efficiency, yielding a positive t-statistic of 8.00 and a p-value of 0.015, indicating a statistically significant difference. Regarding fluency, no significant difference was observed between the two methods (t-statistic =-3.5, p-value = 0.074), suggesting comparable performance in ensuring the natural flow of text. Discussion: ChatGPT-4o showed competitive performance in English-to-Arabic post-editing, particularly in producing fluent, coherent, and stylistically consistent text. Its conversational design enables efficient and consistent editing across various domains. Nonetheless, the model faced challenges in handling grammatical and syntactic nuances, domain-specific idioms, and complex terminology, especially in medical and sports contexts. Overall, the study highlights the potential of ChatGPT-4o as a supportive tool in translation postediting workflows, complementing human translators by enhancing productivity and maintaining acceptable quality standards.

This study presents a comprehensive linguistic analysis of AI prompt phraseology, focusing on the roles of synonymous prompt phrases and collocational prompt phrases in shaping the responses of large language models (LLMs). Using a dataset of 300 prompts tested across ChatGPT, Claude, and Gemini, the research investigates how subtle variations in wording influence the structure, tone, and effectiveness of AI-generated outputs. Results show that while synonymous phrases lead to stylistic and rhetorical variation, collocational phrases trigger more stable and structured responses due to their frequent appearance in training data. The study bridges applied linguistics and prompt engineering, providing theoretical insights and practical strategies for more precise and predictable human-AI interaction. Recommendations are offered for AI users, educators, and developers, emphasizing the strategic role of phraseological awareness in prompt design. The findings contribute to a growing interdisciplinary effort to optimize communication with AI systems through linguistically informed input strategies.

L'implementazione dei Large Language Models (LLM) ha rappresentato un avanzamento significativo nel campo dell'intelligenza artificiale (IA), trasformando radicalmente l'interazione tra esseri umani e sistemi artificiali. Questi modelli avanzati, basati su reti neurali profonde, sono addestrati su enormi corpora testuali che consentono loro di raggiungere i livelli di comprensione, generazione e traduzione linguistica precedentemente ritenute esclusive della cognizione umana 1. Nonostante i vantaggi apportati, l'impiego dei LLM solleva questioni etiche e tecniche di notevole rilevanza, in particolare per la riproduzione di bias culturali e sociali, tra cui quello di genere che emerge come uno dei fenomeni più evidenti e problematici. Questo contributo intende analizzare le origini dei bias nei LLM, fornire esempi concreti della loro manifestazione e proporre approcci di mitigazione degli effetti.

This article provides an epistemological analysis of current attempts of explaining how the relatively simple algorithmic components of neural language models (NLMs) provide them with genuine linguistic competence. After introducing the Transformer architecture, at the basis of most of current NLMs, the paper firstly emphasizes how the central question in the philosophy of AI has been shifted from "can machines think?", as originally put by Alan Turing, to "how can machines think?", pointing to an explanatory gap for NLMs. Subsequently, existing explanatory strategies for the functioning of NLMs are analyzed to argue that they, however debated, do not di er from the explanatory strategies used in cognitive science to explain intelligent behaviors of humans. In particular, available experimental studies turned to test the theory of mind, discourse entity tracking, and property induction in NLMs are examined under the light of the functional analysis in the philosophy of cognitive science; the so-called copying algorithm and the induction head phenomenon of a Transformer are shown to provide a mechanist explanation of in-context learning; finally, current pioneering attempts to use NLMs to predict brain activation patterns when processing language are here shown to involve what we call a co-simulation, in which a NLM and the brain are used to simulate and understand each other.

Recent advances in Large Language Models (LLMs) have enabled their application to recommender systems (RecLLMs), yet concerns remain regarding fairness across demographic and psychological user dimensions. We introduce FairEval, a novel evaluation framework to systematically assess fairness in LLM-based recommendations. Unlike prior benchmarks that focus solely on demographic attributes, FairEval uniquely integrates personality traits with eight sensitive demographic attributes, including gender, race, and age enabling a comprehensive and nuanced assessment of user-level bias. We evaluate state-of-the-art models, including ChatGPT 4o and Gemini 1.5 Flash, on music and movie recommendation tasks using structured prompts. FairEval's personality-aware fairness metric, PAFS, achieves high consistency scores up to 0.9969 for ChatGPT 4o and 0.9997 for Gemini 1.5 Flash, underscoring its robustness in equitable recommendations across diverse user profiles, while also uncovering fairness gaps, with SNSR disparities reaching up to 34.79%. Our results also reveal disparities in recommendation consistency across user identities and prompt formulations, including typographical and multilingual variations. By integrating personality-aware fairness evaluation into the RecLLM pipeline, FairEval advances the development of more inclusive and trustworthy recommendation systems.

CCS Concepts • Information systems → Recommender systems; Personalization; • Human-centered computing → User models; • Computing methodologies → Natural language processing.

Recent advancements in Large Language Models(LLMs) have been notable, yet widespread enterprise adoption remains limited due to various constraints. This paper examines bias in LLMs-a crucial issue affecting their usability, reliability, and fairness. Researchers are developing strategies to mitigate bias, including debiasing layers, specialized reference datasets like Winogender and Winobias, and reinforcement learning with human feedback (RLHF). These techniques have been integrated into the latest LLMs. Our study evaluates gender bias in occupational scenarios and gender, age, and racial bias in crime scenarios across four leading LLMs released in 2024: Gemini 1.5 Pro, Llama 3 70B, Claude 3 Opus, and GPT-4o. Findings reveal that LLMs often depict female characters more frequently than male ones in various occupations, showing a 37% deviation from US BLS data. In crime scenarios, deviations from US FBI data are 54% for gender, 28% for race, and 17% for age. We observe that efforts to reduce gender and racial bias often lead to outcomes that may over-index one sub-class, potentially exacerbating the issue. These results highlight the limitations of current bias mitigation techniques and underscore the need for more effective approaches.

This study evaluates the biases in Gemini. Flash Experimental, a state-ofthe-art large language model (LLM) developed by Google, focusing on content moderation and gender disparities. By comparing its performance to ChatGPTo, examined in a previous work of the author, the analysis highlights some di erences in ethical moderation practices. Gemini. demonstrates reduced gender bias, notably with female-specific prompts achieving a substantial rise in acceptance rates compared to results obtained by ChatGPT-o. It adopts a more permissive stance toward sexual content and maintains relatively high acceptance rates for violent prompts (including gender-specific cases). Despite these changes, whether they constitute an improvement is debatable. While gender bias has been reduced, this reduction comes at the cost of permitting more violent content toward both males and females, potentially normalizing violence rather than mitigating harm. Male-specific prompts still generally receive higher acceptance rates than female-specific ones. These findings underscore the complexities of aligning AI systems with ethical standards, highlighting progress in reducing certain biases while raising concerns about the broader implications of the model's permissiveness. Ongoing refinements are essential to achieve moderation practices that ensure transparency, fairness, and inclusivity without amplifying harmful content.

Diplomsko delo se ukvarja s samoreprezentacijo uporabnikov na aktualni mobilni platformi Tinder s poudarkom na razlikovanju samopredstavljanja med spoloma. Tinder velja danes za eno izmed najbolj razširjenih in inovativnih načinov sklepanja partnerskih razmerij. V raziskavi se osredinjamo na teorijo J. Kilbourne o reprezentaciji spola v oglaševanju, in izpostavljamo eno izmed njenih glavnih ugotovitev podrejenosti žensk na družbenem, seksualnem, čustvenem in izobraževalnem področju. V primerjavi z moškimi so ženske v oglasih reprezentirane kot podrejene moškemu in kot objektivizirana, seksualizirana telesa. Moški pa so v oglasih reprezentirani skozi dejavno in/ali agresivno plat ter izkazujejo premoč nad ženskami. Zanima nas, ali se oz. na kakšen način se podrejenost žensk v primerjavi z moškimi kaže skozi samopredstavljanje na Tindru, in ali moški izkazujejo premoč nad njimi. V empiričnem delu raziskave smo zajeli 10 ženskih in 10 moških uporabnikov omenjene platforme. Analizirali ...

This master's thesis represents an interdisciplinary approach to understanding gender bias manifested in the output of artificial intelligence tools, which are based on language models. Biases and stereotypes become problematic when we systematically exhibit unfair positive or negative bias towards a particular group, leading to heuristics and wrong and unfair decision-making. This work sheds light on how bias in AI tools arises in the first place through a series of steps: first, it explains the methodologies of natural language processing systems and the modelling of word meaning through word embeddings. It then examinesu how gender bias manifests in the language and how language present in a given training dataset can influence the results of word embeddings. In addition to the choice of training datasets, the work also mentions labelling, input data representation, models and research conceptualisation as reasons for gender bias in AI tools. Based on our own studies, published between 2019 and 2021, the thesis shows concrete ways gender bias manifests itself in natural language processing tools. The master's thesis also deals with gender bias from the perspective of human interaction with biased AI tools. Anthropomorphisation and the apparent objectivity of artificial intelligence are one of the main ways in which such tools can negatively influence human decisionmaking. The main contribution of this thesis is to propose, through a broad literature review of sources from linguistics, computer science, philosophy and other disciplines, as well as our own studies, several guidelines that could reduce gender bias in tools such as large language models. Firstly, gender bias needs to be precisely defined with the help of social sciences. At the same time, AI tools must meet high ethical standards, be inclusive and start representing different experiences. A clear framework of accountability for tool developers needs to be established. Companies must commit to continuous improvement of tools and to transparency, even if these do not benefit the company financially. Despite the usefulness of tools that mimic humans, I believe it would be beneficial for developers to commit to reducing the anthropomorphisation of tools, since the latter can undesirably influence decision-making and thus interfere with personal autonomy. One of the guidelines proposed in the paper is the prevention of the so-called "feedback loop phenomenon" that occurs when AI-generated texts are fed as training dataset for further tools. Lastly, the master's thesis proposes to introduce education in AI and gender bias, which can empower people both in their use of tools and as individuals living in the new reality.

Predmetno magistrsko delo predstavlja interdisciplinarni pristop k razumevanju spolne pristranosti, ki se pojavlja v rezultatih orodij umetne inteligence, ki temeljijo na jezikovnih modelih. Pristranosti in stereotipi postanejo problematični, ko sistematično izkazujemo nepravično pozitivno ali negativno nagnjenost do določene skupine, kar lahko vodi to hevristik in napačnega ter nepravičnega odločanja. Delo preko več korakov osvetljuje, kako do pristranosti v orodjih umetne inteligence sploh pride: najprej pojasni metode sistemov obdelave naravnega jezika ter modeliranje pomena besed preko vektorskih besednih vložitev. Nato se ukvarja s tem, kako se spolna pristranost manifestira v jeziku samem, ter kako jezik neke učne množice podatkov vpliva na rezultate besednih vložitev. Kot razloge za spolno pristranost v orodjih umetne inteligence poleg izbire učnih množic navedem še označevanje, predstavitev vhodnih podatkov, modele ter konceptualizacijo raziskav. Na podlagi lastnih študij, objavljenih med leti 2019 in 2021, naloga pokaže, na kakšne načine orodja obdelave naravnega jezika izkazujejo spolno pristranost. Magistrska naloga se s spolno pristranostjo ukvarja tudi z vidika interakcije človeka s pristranimi orodji. Antropomorfizacija in navidezna objektivnost umetne inteligence predstavljata dva izmed glavnih negativnih vplivov tovrstnih orodij na človekovo odločanje. Ključni doprinos naloge je preko širokega pregleda literature virov jezikoslovja, računalništva, filozofije in drugih ved predlagati več smernic, ki bi lahko zmanjšale spolno pristranost v orodjih, kot so veliki jezikovni modeli. Spolno pristranost je potrebno natančno definirati s pomočjo družboslovnih ved. Orodja umetne inteligence morajo hkrati zadoščati visokim etičnim standardom, biti vključevalna in upoštevati zastopanost različnih izkušenj. Za razvijalce orodij je potrebno vzpostaviti jasen okvir odgovornosti. Podjetja se morajo zavezati k nenehnim izboljšavam orodij in transparentnosti, četudi jim te finančno ne koristijo. Kljub koristim, ki jih lahko imamo od orodja, ki posnema človeka, menim, da je dobro, da se razvijalci zavežejo k zmanjševanju antropomorfizacije orodij, saj le-ta lahko neželeno vpliva na človekovo odločanje in s tem posega v njegovo avtonomijo. S tem, ko orodja sama ustvarjajo besedila, ta pa uporabimo kot učno množico za nadaljnja orodja, ustvarjamo »fenomen povratne zanke«. Ena izmed smernic naloge je tovrsten efekt preprečiti. Nazadnje naloga predlaga vodilno vlogo, ki naj jo ima izobrazba na področju umetne inteligence in spolne pristranosti, saj lahko opolnomoči ljudi tako pri uporabi orodij kot pri življenju v novi realnosti.

The debate on language and gender in Slovenia raises the issue of gender-sensitive language use, which goes far beyond mere equality and inclusiveness in language use. While the establishment of (gender) identity in language is related to the theoretical question of the social construction of gender, ensuring gender equality in language is also a matter that requires social engagement in various fields, especially including various segments of the education system. As the first public debate on this issue – organised by the Faculty of Arts, University of Ljubljana, on 23 October 2018, following attempts to reduce gender inequality in legal documents adopted by the Faculty – showed, the approach to language and gender will necessarily be interdisciplinary. Linguistic considerations call for a shift in emphasis from vocabulary and address code (available resources) to arguments of morphosyntactic form and, in particular, to the symbolic structure of linguistic signs, taking into accou...

Spolna diskriminacija in seksizmi vrst in oblik so kljub nekim splošno družbenim težnjam k enakopravnosti med spoloma še vedno del našega vsakdanjega življenja. V seminarski nalogi smo tako svoje zanimanje usmerili v raziskovanje jezika in spola, ki tudi lahko deluje diskriminatorno. Spraševali smo se kako jezik in njegova raba vplivata na območje spola v družbi. Spraševali smo se torej, kakšna je družbena funkcija jezika in kako ta vpliva na konstrukcijo posameznikove realnosti. V zvezi s tem smo se osredotočili predvsem na slovenščino in ugotavljali ali gre v tem primeru za jezik, ki je spolno občutljiv. Tekom raziskovanja smo slovenski jezik prepoznali kot spolno občutljiv, obenem pa kot jezik, ki je zmožen dati glas ženskam. V začetnem teoretičnem delu smo se najprej posvetili krajšim definicijam pojmov na katerih smo svojo seminarsko nalogo gradili: spolno občutljiv jezik, slovnični in družbeni spol, moška generičnost itd. V drugem delu naloge smo pozornost usmerili na jezik kot orodje, ki služi pri oblikovanju družbene realnosti, posledično pa tudi konstituiranje spolne identitete, dojemanje in vidnost spolov v družbenem sistemu. V empiričnem, zadnjem delu naloge pa smo na podlagi sprejetja sklepa 41 na Filozofski fakulteti želeli preko dotičnega primera umestiti teoretska vprašanja in ugotovitve, ki smo si jih skozi seminarsko nalogo postavljali in odgovarjali. Podali smo krajši oris dogajanja na Filozofski fakulteti, na koncu pa obravnavali tudi buren odziv nekaterih spletnih medijev, ki ga je povzročilo sprejetje sklepa.

Cette recherche s’est fondée sur quelques raisons, entre autres, c’est l’expérience du chercheur sur le silence des étudiants existant  en classe lors d’apprentissage. Il faudrait que, dans la classe de langue, notamment de Civilisation Française II, les étudiants participent activement à l’interaction comme discussion, question-réponse, ou débat. Cette recherche a pour but de décrire l’interaction des étudiants en 3e année au cours de Civilisation Française II et de connaitre les portions de parole chez les étudiants et l’enseignant, et aussi les facteurs qui influencent leur interaction. Cette recherche est qualitative avec la méthode micro-ethnographique, face à environ vingtaine étudiants. L’objet de cette recherche est l’interaction, il faut donc enregistrer le processus d’apprentissage. Pour l’analyser, nous avons utilisé la technique de Brown (BIAS). Les résultats montrent que le professeur domine l’interaction au cours de Civilisation Française II. Cela se voit par  le grand nombre des paroles chez le professeur, surtout sa lecture (Teacher Lecture) et sa réponse (Teacher Responds)

For patriarchy to be successful in keeping itself front and centre in the eyes of the "masses" the definition of our species conforming to the stated central role of the male in creation has to be fully operative on a seemingly "secular" level.  This article explains this phenomenon.

Patriarchy created the life-threatening social disease that 11,000 years of rule inflicted on the whole species. The poisonous mix of phallic fantasy, myth and facts, is traumatic stress that the species may not be able to solve its centuries-old problems. Since patriarchy is perpetrated and perpetuated anew every day through male-bias in 20,000 names a big part of the solution is changing names.

During the 11,000 years patriarchy ruled it embedded male-bias in over 20,000 names thus making language suit their ideology of eternal male rule. Every male-biased name stops our species from seeing reality-as-it-is regarding our own species.  Every bias hides a truth creating ignorance of what actually exists to be seen and known.
 
bias, names, language, history, sociology, patriarchy,

18  Defines what patriarchal radicalization is, and that it was jettisoned into the mind to change the mind from being a thinking organ to being a belief-organ. I show that rights to correct information and factual names about the speech-making species can only be sapient rights.

The first bias establishing power in males, creating belief in male superiority. Then developing language using broad categories to embed more male bias. What using a language with a massive number of names embedding male bias does to people. Changing the language to evidence-based language, including neutral, inclusive and specific names would reduce, and perhaps, eliminate the power of male dominance.

Femic clarifications discuss names that communicate confusing messages on planet earth, good and bad, true and false.  That, language being fluid, embedding morphemes in names can create false 'opposites', support lies, and even reinforce them. That a certain number of names made to satisfy a specific category stating an outright lie can succeed.

Rofemtic Vocabulary restores names that reduce the effect of patriarchy's 10,000-strong list of names having the political agenda of keeping patriarchy on the planet as "the way it is."  It suggests names that create balance, that reduce the list of names with embedded male-bias and names that fill a void in the living of life as thinking-and-feeling beings, etc.

The consequences of renaming fem (wo) +man are discussed in this paper.  Seeding the name 'man' in fem's name deeply affected and changed mind in both genders.  It shrank mind in both genders to believe in illogic, fantasy and nonsense, causing the stress of conflict between fantasy and fact in both genders, not diagnosed to date as mental stress and a form of insanity.

In this paper I try to show the broad field -of living life as minded beings- patriarchy ploughed through to develop language that would make "the masses" believe in their phallic-based fantasy that the male was superior and thus, the 'mind' of the species.  How religion is made to fit this male superiority into the development of language through 11,000 years of patriarchal rule becomes observable.

Discusses the expansion of the direction the Sumerians' set after they named the phallus Creator God in 9000 B.C.E. Names supporting Phallus were invented to create belief in Phallus by naming the roles of the male genitals sacred.  This shows that patriarchy's phallus-supporting House of Names, using male genitalia to create details, make Phallus sound like it is Truth.

This paper looks at the symbols made and used as names for the two members of our species and the name of our species.  The symbols were to be believed to be a theory.  The author examines the names from the perspective of evidence and logic in view of speech-making and speech-using, to see if they fulfill the requirements of being a theory and finds the symbols/names  lacking.